MCP Server Discovery - October 23, 2026 (Evening Sweep)¶
An eight-hour delta after the midday supplement closed, and the productive lane was the submission window again: eighteen issues, #5994 through #6011, read in full. The directories were nearly silent. The mcp.so feed carried thirty cards with exactly one posted after the previous capture, and mcpservers.org /all pages 1 to 3 returned ninety unique slugs against ninety at the supplement, every one already dispositioned once the brand-name slugs were unravelled.
One entry cleared the bar, and it is the first surface in this catalog that prices the marketplace rather than the product: what Etsy, eBay and Amazon actually take, itemised, dated and reproducible.
New Servers Catalogued¶
| Server | Category | What it does |
|---|---|---|
| ShopTrueProfit MCP | Commerce & E-Commerce | Per-unit marketplace economics for Etsy, eBay.com (US) and Amazon.com (US, FBA and FBM): thirteen read-only tools covering fees, profit, margin, break-even price, the price needed for a target margin, and a three-way comparison of the same product. https://mcp.apify.com/?tools=vncntlabs/shoptrueprofit-fee-mcp hosted on Apify Standby, requiring an Apify API token: unauthenticated POST initialize returned HTTP 401 {"error":"invalid_token","error_description":"Missing or invalid access token. Pass an Apify API token in the Authorization: Bearer header."}. The drivable path is the MIT repository PravinrajV/shoptrueprofit-fee-mcp (created 2026-10-17, 0 stars, 68 files), which clones, builds and serves the same thirteen tools with no credential at all: npm install, npm run build, PORT=3311 node dist/http.js, and POST initialize returned HTTP 200 with serverInfo shoptrueprofit-fee-mcp 1.0.0 on protocol 2025-06-18, tools/list returning all thirteen. Driven with five real calls. ebay_profit at price 25, shipping_charged 5, item_cost 8, shipping_label_cost 4, promoted_listings_pct 5 returned "eBay profit $12.02 on $30.00 revenue (40.07% margin) after $5.98 fees and $12.00 costs. Break-even price $10.24 per unit," with three itemised lines carrying their basis: final value fee $4.08 at "13.6% on the portion up to $7,500; 2.35% on the portion over $7,500 (per item, on the total amount of the sale incl. shipping and tax)", per-order fee $0.40 because "order total $30.00 > $10", promoted listings $1.50 as "5% of $30.00." etsy_fees at price 24.99, shipping_charged 5, offsite_ads_pct 15, region us returned "$7.80 (26% of revenue)" across listing $0.20, transaction $1.95 at 6.5%, payment processing $1.15 at "3% of $29.99 + 0.25 USD" and Offsite Ads $4.50 at "15% of $29.99, max 100 per order." amazon_fees at price 29.99, category home-kitchen, fulfillment fba, 10 x 8 x 4 inches, 1.5 lb, ship_date 2026-11-20 returned "$11.66 (38.89% of revenue)": referral $4.50 at 15% with a $0.30 minimum, FBA fulfilment $6.49 priced on "Large standard, 2.3 lb shipping weight, peak standard card" - dimensional weight beating the 1.5 lb entered, because 10 x 8 x 4 inches is 320 cubic inches - and a $0.23 fuel and logistics surcharge that exists only because the ship date sits inside the October 15 to January 14 peak window. compare_marketplaces at price 34.99, item_cost 9, shipping_label_cost 4.5, shipping_charged 5, target_margin_pct 30 returned Etsy $22.24 profit at 55.62% margin, eBay $20.65 at 51.64%, Amazon FBA $13.58 at 38.8%, with break-even $10.42, $11.09 and the Amazon figure following, and the Amazon revenue at $34.99 rather than $39.99 because FBA shipping is set by Amazon and the buyer's $5 never arrives. amazon_target_price at category home-kitchen, fulfillment fba, target_margin_pct 30, item_cost 9 returned "list at $29.39 for a 30% margin; break-even $19.02," with revenue $29.39, fees $11.57, costs $9.00, profit $8.82 and margin 30.00% at that price - solved rather than divided, because the referral percentage and the fulfilment band both move with the price. Reference lookups returned count: 11 eBay final-value-fee categories with the per-order fee rule and seller-performance options, count: 37 Amazon referral categories with rates, minimums and size tiers, and count: 12 Etsy Payments regions with processing rate, flat fee and currency. Every response carries a dated fee table - ebay-us-2026-10-16, amazon-us-2026-10-16, etsy-2026-10-13 - the engine file that produced it and an engine_sha256 for each, the assumptions applied, the sources as official Seller Central and marketplace help-centre URLs, and a calculator_url back to the seller-facing page the same engine drives. Official registry entry io.github.PravinrajV/shoptrueprofit-fee-mcp 1.0.0, active since 2026-10-17T14:12:06Z, published as an mcpb package with its own fileSha256. Gap named rather than claimed: the vendor's strongest accuracy statement is that the suite cross-checks this server against the website's own functions on 1,700+ cases including Amazon's and eBay's own worked examples, and that claim is not reproducible from a public clone - those tests need the private app source and skip themselves. npm test on the clone returned 94 passed and 31 skipped, the entire crosscheck.test.ts file among the skips, which the README states plainly rather than letting a green run imply otherwise. What a reader can run does pass: protocol, schemas, distribution, batch, packaging and an Apify Standby test on the real SDK. Also named in the entry rather than smoothed over: what is deliberately not modelled (eBay Store subscriptions; Etsy Ads cost-per-click, Etsy Plus, currency conversion and the Regulatory Operating Fee; Amazon's low-inventory, inbound-placement, returns-processing, aged-inventory and storage-utilisation surcharges, lithium battery fees, removal and disposal, the Overmax handling fee, and FBA rates after January 14 2027, which are unpublished and fall back to 2026 non-peak rates with a warning), the US-only marketplace scope, and a one-week-old repository with no stars. |
Candidates Read and Held¶
On the mcp.so feed, one card posted after the midday supplement closed at 19:19Z, and it surfaced two listings from a single timestamp, 2026-10-23T21:00:21.864Z:
- AppGoat (
createdAt 2026-10-23T21:00:21.864Z, Data & Analytics,https://api.appgoat.ai/mcp): App Store and Google Play data for agents - apps, reviews, keyword research, top charts, competitors and ASO audits. Probed live:POST initializewithout a key returned HTTP 200 withserverInfoappgoat0.2.0 on protocol 2025-06-18 and both the tools and prompts capabilities,tools/listreturned 29 tools spanningget_app,get_reviews,get_app_history,get_app_changes,get_app_keywords,get_competitors,search_appsover a catalogue of about 5 million apps in both stores,get_charts,keyword_research,watch/get_events,audit_app,get_insights,list_playbooks, pluslist_connections,read_connection,propose_action,apply_action,get_actionandreject_actionover connected App Store Connect, Play Console and Apple Ads accounts with a dry-run diff before any write; andtools/callreturned HTTP 401{"error":"invalid_token","error_description":"Missing or invalid access token"}, so the keyless handshake is a discovery surface and every datum needs an account. Held as class-covered, and the class is dense: Applyra already carries both stores with 25 tools over rank tracking, difficulty and traffic, listing audits, metadata simulation, competitors, autocomplete and top charts; StorePilot carries both stores as portfolio operations with 34 tools over vitals, crash and ANR thresholds, installs, ratings, earnings and reviews; Appbot carries review intelligence across the App Store, Google Play and other stores; ASOgenic carries Apple keyword research, metadata authoring and submission; BestAppify carries 40 tools over the Shopify app store. The write path with an approval diff is the one thing AppGoat adds and it is already carried by StorePilot's App Store Connect reads and ASOgenic's reviewed submission. Recorded for reconsideration if any of those five thins, not opened as a sixth. - Pokemon TCG Restock Alerts carried its midday-supplement hold as consumer collectibles data on the same timestamp, re-listed rather than new.
From the submission window #5994 through #6011, eighteen issues in eight hours. ShopTrueProfit is above; the other seventeen read and held:
- API Status Check (
#6008, keyless athttps://apistatuscheck.com/api/mcp, 8 tools): live and historical availability for public APIs. Driven with two real calls:POST initializereturned HTTP 200 withserverInfoapistatuscheck0.1.0, andlist_down_apisreturnedcount: 9entries atmonitorLastRun 2026-10-24T00:05:00.448Zwith Anthropicdegraded("Partially Degraded Service"), Cursordegraded("Minor Service Outage") and Twiliodegraded("Partially Degraded Service"), each carrying the official status page and a per-API page on the vendor's site. Aget_api_statuscall passingapiinstead of the schema'sslugwas rejected with a field-naming validation error rather than a silent default, which is the right behaviour. Held as developer operations triage rather than an operator data surface: it answers "is the vendor broken or is my code broken", which is a real question and the class this catalog already carries in healthchecks, Visualping's change monitoring and the uptime entries, and it holds no record of the operator's own business. Worth a second look if the catalog ever needs a vendor-outage surface beside its own monitoring. - MyMCPTools (
#6009, keyless athttps://mymcptools.com/api/mcp, 8 tools over a live-verified directory of MCP servers with handshake probes, probe history, reconstructed outage incidents and schema drift): driven toinitialize200 withserverInfomymcptools1.0.0. Held as directory infrastructure - it is a directory of the thing this catalog is a catalog of, and the schema-drift monitoring it does is worth knowing exists without being an operator surface. - AISOTools (
#6010, keyless athttps://aisotools.com/api/mcp, 6 tools,serverInfoaisotools0.2.0, a directory of 5,780 AI tools pluscheck_ai_visibility): held as directory infrastructure and against the saturated AI-visibility class, the same disposition WebGlazer and the other visibility entries carry. - RatedWithAI Accessibility Checker (
#6007, keyless athttps://ratedwithai.com/api/mcp, 2 tools,serverInforatedwithai1.0.0): axe-core scans in a headless browser returning failing rules, affected element counts and example markup, with an explicit statement that it does not certify compliance. Held as web quality tooling, the class carried by squirrelscan; the honesty in the description is exactly right and does not change the class. - Context Atlas (
#6002, keyless athttps://contextatlas.dev/mcp, 3 tools): a catalogue of the sources vendors publish for coding agents - llms.txt, docs, MCP servers, skills, rule files, CLIs - grouped by tool with detection signals. Probed toinitialize200,serverInfocontextatlas1.0.0. Held as coding-agent developer tooling, a class the catalog covers in its own docs and skills surfaces. - SlashYear (
#5998, keyless athttps://slashyear.com/mcp, 11 tools,serverInfoslashyear1.2.0): history by year, date and country, with every line quoted verbatim from a specific English Wikipedia revision carrying that revision id. The sourcing discipline is the best in the window and the subject is reference knowledge, not an operator's business record - held as general knowledge. - A4A (
#6000, local stdio,pip install a4a-mcp, cryptographicapub1addressing over NIP-44 and NIP-59 with geo-distributed relays and a default-deny quarantine for cross-boundary prompt injection): carried its October 22 hold as agent messaging infrastructure, re-read only because it arrived again as a fresh submission. - Penthera (
#5999, local stdio, MIT, offline rule-based repo, dependency, agent-config and runtime security checks with file:line evidence and no shell execution): held as developer security tooling, which this catalog treats as tooling rather than a data surface even when it is unusually well-mannered about read-only guarantees. - VibeView (
#5994, local stdio, cloud iOS, Android, Apple TV, Android TV and Roku devices for coding agents with tap, type, swipe, UI tree, screenshots and logs): held as device-farm developer infrastructure beside Zeptap and optics-framework. - Kaption WhatsApp (
#6004, remote OAuth with PKCE and dynamic registration plus a local stdio package, 17 tools over chat search, contacts, groups, voice-note transcripts, labels, reminders and scheduled messages, with no send tool): held against the dense messaging and mailbox class; the read-only discipline and the "no tool sends a message" statement are the strongest arguments it has and do not open a new class. - ViralRecast (
#6003,https://viralrecast.com/mcp, OAuth 2.1 with dynamic registration and PKCE or a bearer key, 18 tools over TikTok, Instagram, YouTube, X, Threads, LinkedIn and Reddit post content with transcripts, carousel slides, top comments, metrics, a Viral Score relativised to the author's size, trending by niche, creator outlier posts, the ads a competitor runs on Meta, Google, TikTok and LinkedIn with their running days, and brand mentions): the competitor-ads block with days-running is the most interesting element and the class is carried by the ad-intelligence entries, Adrails above all, plus the social-listening entries. Held; recorded as a revisit candidate if the ad-tracking depth holds up beyond a submission description. - AISTLabs (
#6001,https://aistlabs.ai/mcp, OAuth 2.1 or bearer key): creative generation across video, image and audio models with credits quoted before spend. Held as media generation, the class carried by Mage, AIKIZI, Meshy and the rest. - BestJapanBuy (
#6006,https://mcp.bestjapanbuy.com/mcp, OAuth 2.1 with a free/api/v1beside it, 8 tools over Japanese product prices, China-compatibility checks for voltage and plugs, store locations and the November 2026 tax-refund rules): held as consumer travel and shopping data, a strong vertical and not an operator's system of record. - pptx-mcp (
#6005, local stdio via npx): read, outline and preview PowerPoint files. Held as a document utility, the class carried by the docx, pdf and office entries. - The three job-board servers (
#5995mcp-elempleo,#5996mcp-magneto,#5997mcp-workatastartup, all from one author, all local stdio, all scaffolded around a local "master profile" with an explicit confirmation step before applying): held as job-search automation for individuals, and the near-identical shape filed inside five minutes is the single-author burst pattern this catalog excludes whole rather than cherry-picking.
On mcpservers.org /all pages 1 to 3, ninety unique slugs against ninety at the midday supplement, and every apparent newcomer resolved to a prior disposition once the URL-shaped slugs were matched to their brands: www-reportgem-com-mcp-html is ReportGem, catalogued on an earlier pass and probed again here (keyless POST initialize 200 with the tools capability and five search tools - search_research, search_filings, search_meetings, search_visuals, get_evidence); lfg-gm-agents is LFG Alarm, carried in the October 22 night holds (probe this cycle: POST initialize returned HTTP 401 {"jsonrpc":"2.0","error":{"code":-32001,"message":"Connect LFG Alarm to sign in."},"id":null}); medicine-shanchuangkj-com-mcp is MedicineData, sitepatent-com-mcp is SitePatent, twinmind-com-blogs-twinmind-mcp is TwinMind, winelabs-ai-agents is Wine Labs, planlimits-rpatel021-vercel-app-mcp-docs is PlanLimits, terminal-tax-integrations is Terminal Tax, underanchor-com-docs-connect-ai is Underanchor, uain-global-for-ai is UAIN, sitenet-api-openyuri-workers-dev-connect is the OpenYuri Tools Network, prosignature-co-for-ai-agents is ProSignature, shipday-public-data-...-documentation2-pdf is Shipday and wrdo-co-za is WRDO. Two genuinely new slugs were dead ends rather than candidates: www-bubblav-com is listed as a live card while its endpoint answers HTTP 404 {"error":{"code":"404","message":"The deployment could not be found on Vercel."}}, and my-websearch serves a 404 detail page.
Source Behaviour Worth Recording¶
A listing is not a deployment, and this cycle produced both halves of the failure. www-bubblav-com sits in the directory as a product page for a support-chat platform while the endpoint its own page implies returns a Vercel deployment-not-found. my-websearch is a slug the directory's sitemap knows and its detail route does not. Neither is a catalog candidate, and neither is visible to anyone who reads the directory rather than probes it.
A keyless handshake is not a usable server. AppGoat answered initialize and tools/list with no credential at all - 29 tools enumerated, prompts advertised - and then returned 401 invalid_token on the first tools/call. PourWire, catalogued in the midday supplement, is the opposite case: keyless all the way through and drivable. The two look identical on a card and at handshake, and only a tools/call separates them. That is now three cycles in a row where the probe, not the listing, decided the entry.
A directory slug is a URL, not a name. Half of this cycle's apparent newcomers were the same servers under their documentation URLs. Matching on the slug alone would have produced thirteen false positives in a single pass. The brand token in the slug, matched against the catalog's own text, is the only comparison that has held.
Method¶
Sources: mcp.so /feed page 1 read from the TanStack payload with a browser user agent (30 unique createdAt values, newest 2026-10-23T21:00:21.864Z, oldest 2026-10-22T00:18:16.464Z); mcpservers.org /all pages 1 to 3 newest-first through the r.jina.ai reader proxy (90 unique slugs, each resolved to a prior disposition); the chatmcp/mcpso issue window #5994 through #6011 read over the REST API with full bodies; mcp.so detail page for AppGoat; and live JSON-RPC probes against nine endpoints (mymcptools.com, apistatuscheck.com, ratedwithai.com, contextatlas.dev, slashyear.com, aisotools.com, api.appgoat.ai, www.reportgem.com, mcp.lfg.gm, mcp.bubblav.com, plus the Apify Standby URL for ShopTrueProfit). ShopTrueProfit was settled on a local build of the MIT repository - clone, npm install, npm run build, HTTP server on 127.0.0.1:3311, initialize, tools/list and five tools/call executions - after the hosted endpoint returned 401, with npm test run to record 94 passed and 31 skipped, the fee tables read from the responses themselves (ebay-us-2026-10-16, amazon-us-2026-10-16, etsy-2026-10-13, each with its engine_sha256), and the official registry entry io.github.PravinrajV/shoptrueprofit-fee-mcp verified through registry.modelcontextprotocol.io.
Catalog 981 to 982 servers (+861 to +862 guides).